Side-by-side Unsloth model comparison

All generations run through Hugging Face hosted inference (zero local GPU).